Flood and storm forecasting is a critical issue in Vietnam, a country that is frequently affected by natural disasters. With the development of artificial intelligence, particularly deep learning, the construction of accurate forecasting systems has become more feasible. In this study, we propose the use of deep learning models such as LSTM, Transformer, and Temporal Convolutional Network (TCN) to analyze and predict floods and storms based on meteorological and hydrological data. The input data is collected from OpenDevelopmentMekong. The accuracies of the models are as follows: LSTM achieves 0.9593 (Storms), 0.9576 (Floods); Transformer achieves 0.9657 (Storms), 0.9618 (Floods); and Temporal Convolutional Network achieves 0.9754 (Storms), 0.9742 (Floods). The deep learning models are optimized to handle spatiotemporal data, enabling accurate predictions of storm and flood developments. The results of the study show that this method significantly improves accuracy compared to traditional models. The system is integrated with GIS tools for visualization and early warning support. This research not only contributes to improving flood and storm forecasting capabilities in Vietnam but also lays the foundation for applying AI in disaster prevention and management.